SYNTOLOGY HomeExplorerAtlasCodeMethodologyAboutDevelopersFeedPricing
Paper · 2101.08596 · ICLR · 2021

LEAF: A Learnable Frontend for Audio Classification

Olivier Teboul, Neil Zeghidour, Félix De Chaumont Quitry, Marco Tagliasacchi

arXiv · PDF · Open in the Atlas

Code that ran

14 of 26 samples (functions and classes) lifted from this paper's linked repositories executed on a synthesized input in Syntology's jail (1 of them constructed an object rather than computing a result; 1 honoured the contract we drafted, 2 violated it and 4 have no contract verdict; 7 failures of our instrument; 12 no recorded run), measured 2026-11-21; repository links from the Papers with Code archive attributed under CC BY-SA, no affiliation; bodies inline only under a permissive licence as GitHub detects it. Not a reproduction of the paper's results.

"Ran" means the function executed on a synthesized input and returned a value. It is not a reproduction of the paper's results. This record, dated, as JSON: /api/ran/2101.08596.json · badge: /api/ran/2101.08596.svg

Each repository's role is a label: a classifier's output recorded on the link, or, where no role is recorded, what the Papers with Code archive marked. The line under it names its basis. Neither says who wrote the code.

RepositoryRoleRan
SarthakYadav/audax community
Papers with Code archive
8 ran · 10 no recorded run
cpjku/efficientleaf community
Papers with Code archive
5 ran · 1 no recorded run
SarthakYadav/leaf-pytorch community
Papers with Code archive
1 ran · 1 no recorded run
FunctionStatusWhere it lives
GaborConstraint Ran
ran on a synthesized input
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("6e985a21ec9fa940")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned ArrayImpl, ndim 2, dtype family float, shape (2, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import math
import jax.numpy as jnp
from jax import numpy as jnp, lax

class GaborConstraint:

    def __init__(self, kernel_size: int):
        self.kernel_size = kernel_size

    def __call__(self, kernel):
        mu_lower = 0.0
        mu_upper = math.pi
        sigma_lower = 4 * math.sqrt(2 * math.log(2)) / math.pi
        sigma_upper = self.kernel_size * math.sqrt(2 * math.log(2)) / math.pi
        clipped_mu = jnp.clip(kernel[:, 0], mu_lower, mu_upper)
        clipped_sigma = jnp.clip(kernel[:, 1], sigma_lower, sigma_upper)
        return jnp.stack([clipped_mu, clipped_sigma], axis=1)


INIT_ARGS = {
    'kernel_size': 16
}

ARGS = {
    'kernel': jnp.array([
        [1.5, 0.5],
        [2.0, 1.2],
    ])
}

if __name__ == '__main__':
    constraint = GaborConstraint(**INIT_ARGS)
    result = constraint(**ARGS)
    print("Result shape:", result.shape)
    print("Result:", result)
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (2, 2) — shape (2, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
GaborFilterbank Ran
ran on a synthesized input
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("a2fcbfa9f777c17f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

# INIT_ARGS for GaborFilterbank.__init__
INIT_ARGS = {
    'n_filters': 8,
    'min_freq': 50.0,
    'max_freq': 4000.0,
    'sample_rate': 16000,
    'filter_size': 31,
    'pool_size': 15,
    'pool_stride': 5,
    'pool_init': 0.4
}

# Create input tensor for forward pass
# Shape: (batch_size, in_channels, signal_length)
# For 1D convolution, in_channels should be 1 (single channel audio)
batch_size = 2
signal_length = 100

x = torch.randn(batch_size, 1, signal_length)

# ARGS for forward method
ARGS = {
    'x': x
}
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float — no output shape is stated: observe_signature.py observed this class with its constructor sizes canonicalised where it could, which the node does not record, so its shape is not stated as this setup block's
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
PCEN Ran
ran on a synthesized input
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("3aaf38be4c3fb084")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, class construct-then-call prompt, model not recorded) and returned Tensor, ndim 3, dtype family float, shape (2, 4, 8) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import numpy as np
import torch
import torch.nn as nn

INIT_ARGS = {
    'num_bands': 4,
    's': 0.025,
    'alpha': 1.0,
    'delta': 1.0,
    'r': 1.0,
    'eps': 1e-06,
    'learn_logs': True,
    'clamp': None,
}

ARGS = {
    'x': torch.randn(2, 4, 8),
}

if __name__ == '__main__':
    from typing import Optional
    
    class PCEN(nn.Module):
        """
        Trainable PCEN (Per-Channel Energy Normalization) layer:
        .. math::
            Y = (\\frac{X}{(\\epsilon + M)^\\alpha} + \\delta)^r - \\delta^r
            M_t = (1 - s) M_{t - 1} + s X_t

        Args:
            num_bands: Number of frequency bands (previous to last input dimension)
            s: Initial value for :math:`s`
            alpha: Initial value for :math:`alpha`
            delta: Initial value for :math:`delta`
            r: Initial value for :math:`r`
            eps: Value for :math:`eps`
            learn_logs: If false-ish, instead of learning the logarithm of each
              parameter (as in the PCEN paper), learn the inverse of :math:`r` and
              all other parameters directly (as in the LEAF paper).
            clamp: If given, clamps the input to the given minimum value before
              applying PCEN.
        """

        def __init__(self, num_bands: int, s: float=0.025, alpha: float=1.0, delta: float=1.0, r: float=1.0, eps: float=1e-06, learn_logs: bool=True, clamp: Optional[float]=None):
            super(PCEN, self).__init__()
            if learn_logs:
                s = np.log(s)
                alpha = np.log(alpha)
                delta = np.log(delta)
                r = np.log(r)
            else:
                r = 1.0 / r
            self.learn_logs = learn_logs
            self.s = nn.Parameter(torch.full((num_bands,), float(s)))
            self.alpha = nn.Parameter(torch.full((num_bands,), float(alpha)))
            self.delta = nn.Parameter(torch.full((num_bands,), float(delta)))
            self.r = nn.Parameter(torch.full((num_bands,), float(r)))
            self.eps = torch.as_tensor(eps)
            self.clamp = clamp

        def forward(self, x):
            if self.clamp is not None:
                x = x.clamp(min=self.clamp)
            if self.learn_logs:
                s = self.s.exp()
                alpha = self.alpha.exp()
                delta = self.delta.exp()
                r = self.r.exp()
            else:
                s = self.s
                alpha = self.alpha.clamp(max=1)
                delta = self.delta.clamp(min=0)
                r = 1.0 / self.r.clamp(min=1)
            alpha = alpha[:, np.newaxis]
            delta = delta[:, np.newaxis]
            r = r[:, np.newaxis]
            smoother = [x[..., 0]]
            for frame in range(1, x.shape[-1]):
                smoother.append((1 - s) * smoother[-1] + s * x[..., frame])
            smoother = torch.stack(smoother, -1)
            smoother = torch.exp(-alpha * (torch.log(self.eps) + torch.log1p(smoother / self.eps)))
            return (x * smoother + delta) ** r - delta ** r

    pcen = PCEN(**INIT_ARGS)
    output = pcen(**ARGS)
    print(f"Output shape: {output.shape}")
    print(f"Output: {output}")
Setup block drafted by
a language model, class construct-then-call prompt, model not recorded
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (2, 4, 8) — shape (2, 4, 8) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: torch and numpy seeded to 0 by the harness immediately before the class was constructed from INIT_ARGS (fixture_check.py)
SquaredModulus Ran
ran on a synthesized input · constructed an object; no result was computed
SarthakYadav/leaf-pytorch/leaf_pytorch/frontend.py
pointer only (licence: NONE) · get_code("889fadfc65f3beb7")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, SarthakYadav/leaf-pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. A run is recorded.

constant_init Ran
ran; our contract draft was wrong, not the code
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("9f0244a02ae7fba6")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ArrayImpl, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax
from jax import numpy as jnp
from jax import dtypes

# Create a simple PRNGKey for JAX
key = jax.random.PRNGKey(0)

# Define shape as a tuple of integers
shape = (2, 4)

# Define dtype - using jnp.float32
dtype = jnp.float32

# Define constant value
constant = 0.04

ARGS = {
    "key": key,
    "shape": shape,
    "dtype": dtype,
    "constant": constant,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gabor_filters Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("e04f1c11885e4803")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ArrayImpl, ndim 2, dtype family float, shape (4, 401) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax.numpy as jnp
import math

# Construct kernel argument
# kernel should be a 2D array where:
# - kernel[:, 0] contains center frequencies
# - kernel[:, 1] contains fwhm (full width at half maximum) values
# Shape: (num_filters, 2)
kernel = jnp.array([
    [0.5, 0.1],
    [1.0, 0.2],
    [1.5, 0.15],
    [2.0, 0.25]
], dtype=jnp.float32)

# size parameter - kernel size for the impulse response
size = 401

ARGS = {
    "kernel": kernel,
    "size": size
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 2, dtype family float, shape (4, 401) — shape (4, 401) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gabor_filters Ran
ran; our contract draft was wrong, not the code
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("6a7777a3bb79d705")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 2, dtype family float, shape (4, 16) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Define arguments for gabor_filters function
size = 16
center_freqs = torch.tensor([0.1, 0.2, 0.3, 0.4], dtype=torch.float32)
sigmas = torch.tensor([0.5, 0.6, 0.7, 0.8], dtype=torch.float32)

ARGS = {
    "size": size,
    "center_freqs": center_freqs,
    "sigmas": sigmas,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 2, dtype family float, shape (4, 16) — shape (4, 16) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
gauss_windows Ran
ran; our contract draft was wrong, not the code
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("660642538cf0749b")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned Tensor, ndim 3, dtype family float, shape (4, 1, 5) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Construct arguments for gauss_windows
size = 5
sigmas = torch.tensor([[0.5], [1.0], [1.5], [2.0]], dtype=torch.float32)

ARGS = {
    "size": size,
    "sigmas": sigmas,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type Tensor, ndim 3, dtype family float, shape (4, 1, 5) — shape (4, 1, 5) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
hz_to_leaf_param Ran
ran, violated the contract we drafted
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("cd18f02a2caf254f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ArrayImpl, ndim 1, dtype family float, shape (4,) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax.numpy as jnp

# Construct arguments for hz_to_leaf_param
freqs_hz = jnp.array([100.0, 200.0, 400.0, 800.0])
sample_rate = 16000

ARGS = {
    "freqs_hz": freqs_hz,
    "sample_rate": sample_rate,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 1, dtype family float, shape (4,) — shape (4,) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
leaf_truncated_init Ran
ran; our contract draft was wrong, not the code
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("0e0e40f60cb56b3c")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax
import jax.numpy as jnp
import math

ARGS = {
    "dtype": jnp.float32,
    "kernel_size": 401,
    "min_freq": 0.0,
    "max_freq": 8000.0,
    "sample_rate": 16000,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
leaf_uniform_init Ran
ran; our contract draft was wrong, not the code
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("c1b960ad4b82d46e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned function. The image it ran in was not recorded. If your copy returns the same type on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax
import jax.numpy as jnp
import math

ARGS = {
    "dtype": jnp.float32,
    "kernel_size": 401,
    "min_freq": 0.0,
    "max_freq": 8000.0,
    "sample_rate": 16000,
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type function — a function output has no array shape of its own; none is stated — the instrument coerces any output to an array; for a function output the recorded ndim 0 and dtype family float describe that coercion, not the output, and are not stated
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
mel_filter_params Ran
ran, honoured the contract we drafted
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("787e3190e5d34e8e")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned tuple, ndim 2, dtype family float; no output shape is stated. The image it ran in was not recorded. If your copy returns the same type and ndim on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

ARGS = {
    "n_filters": 10,
    "min_freq": 50.0,
    "max_freq": 8000.0,
    "sample_rate": 16000
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type tuple, ndim 2, dtype family float — a tuple output has no array shape of its own; none is stated
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
simple_min_max_scale Ran
ran, violated the contract we drafted
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("c86acac67c0dd466")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ndarray, ndim 2, dtype family float, shape (2, 4) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import torch
import numpy as np

torch.manual_seed(0)

# Create a small numpy array with shape (2, 4) - batch size 2, 4 features
x = np.array([
    [1.0, 2.0, 3.0, 4.0],
    [5.0, 6.0, 7.0, 8.0]
], dtype=np.float32)

minval = 0.0
maxval = 1.0

ARGS = {
    "x": x,
    "minval": minval,
    "maxval": maxval
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
numpy, torch — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ndarray, ndim 2, dtype family float, shape (2, 4) — shape (2, 4) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
squared_modulus_activation Ran
ran; the contract check could not run (RAISES, DEP_MISSING or TIMEOUT)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("7089f92c384fd85f")
How it ran

Ran in Syntology's sandbox on this synthesized input (drafted by a language model, Claude Haiku 4.5, prompt python_fixture_v1) and returned ArrayImpl, ndim 3, dtype family float, shape (2, 8, 2) as a later run recorded it. The image it ran in was not recorded. If your copy returns the same type, ndim and shape on these inputs, it matched what we observed; nothing here checks that. It does NOT mean the output is correct: nothing has compared it to the paper.

import jax
import jax.numpy as jnp
from flax import linen as nn

# Create a small input tensor with shape (batch=2, channels=4, length=8)
# The function expects inputs with shape (batch, length, channels) after transpose
# So we create (batch=2, length=8, channels=4)
inputs = jnp.ones((2, 8, 4), dtype=jnp.float32)

ARGS = {
    "inputs": inputs
}
Setup block drafted by
a language model, Claude Haiku 4.5, prompt python_fixture_v1
Imports
flax, jax — import names as written, not package names to install; a name the repository itself provides is not a package
Returned
type ArrayImpl, ndim 3, dtype family float, shape (2, 8, 2) — shape (2, 8, 2) as a later run recorded it
dtype family
int means an integer dtype; float means any other dtype the instrument saw, bool and complex included (fixture_check.py records only these two families)
Environment
image not recorded; re-verify against syntology-harvest-sandbox:v5-deps; not recorded for this sample: the harness installs a cuda-to-cpu shim (cuda_shim.py) before every sample and no run records whether a sample needed it; seed: none by the harness; any seed is in the setup block
ExponentialMovingAverage Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("b68e7781b9a3baa7")
How it ran

no invocation recorded.

GaborConv1D Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("edbd1816bc9a31c9")
How it ran

no invocation recorded.

GaborFilterHelper Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("63085169c36eb6e3")
How it ran

no invocation recorded.

GaborInit Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("5c70b74778f9633c")
How it ran

no invocation recorded.

GaussianLowPassPooling Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("869cfc21734ba1d4")
How it ran

no invocation recorded.

Leaf Not yet run
unverified (harvested, no recorded run)
SarthakYadav/leaf-pytorch/leaf_pytorch/frontend.py
pointer only (licence: NONE) · get_code("1e7e303c902bf303")
How it ran

The body of this sample is withheld: GitHub detected no licence in this paper's repository, SarthakYadav/leaf-pytorch when it was checked. The setup block that called it is withheld with it, because it is written against that body. source_path and origin say where the code lives, to read under the repository's own terms. No invocation recorded.

Leaf Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("d115cb81f2c624af")
How it ran

no invocation recorded.

Leaf Not yet run
unverified (harvested, no recorded run)
cpjku/efficientleaf/model/leaf.py
code served (permissive licence) · get_code("4507960a65ca7778")
How it ran

no invocation recorded.

PCENLayer Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("cf242b9bc664970f")
How it ran

no invocation recorded.

gabor_impulse_response Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("a86dc35443b7ddfd")
How it ran

no invocation recorded.

gabor_init Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("5fc0b90b915410cb")
How it ran

no invocation recorded.

gaussian_lowpass_kernel Not yet run
unverified (harvested, no recorded run)
SarthakYadav/audax/audax/frontends/leaf.py
code served (permissive licence) · get_code("3201fcfbadfc8198")
How it ran

no invocation recorded.

Also present in other papers' code

Of 26 samples lifted from this paper's linked repositories, 26 carry a function identity and 0 do not. Of the 26 distinct functions with an identity, 1 are also present in other papers whose code we hold (0 with at least one paper dated earlier, 0 the same month, 1 later) and 25 are only here among the 52,977 papers whose code Syntology holds (as of 2026-09-26).

Only here means: among the 52,977 papers whose code Syntology holds (51,887 with lifted samples, 1,090 through their assemblies only), as of 2026-09-26, no other paper's lifted samples or current assembly bricks contain an identical function body after normalisation.

FunctionAlso present inEarliest-datedOthers named
constant_init
brick of this paper's assembly: helper
1 other paper
1 later
2201.03545 (2022-01)
A ConvNet for the 2020s

Presence, not direction: an identical function body in two papers' repositories does not say who wrote it, which came from which, or that either paper's method rests on it. Matching is exact after normalisation, so a renamed or edited function is not matched. Dates are when papers appeared, not when code was written. Only .py files the harvester lifted are covered; notebooks are not read. A new-style arXiv id YYMM.NNNN dates a paper to the month 20YY-MM; otherwise Paper.year dates it to the year; otherwise it is undated. Dates are when the paper appeared, not when the code was written. Read 2026-11-21.

Repositories linked to this paper

Some links come from the archived Papers with Code dataset (CC BY-SA 4.0): attribution and licence.

Abstract

Mel-filterbanks are fixed, engineered audio features which emulate human perception and have been used through the history of audio understanding up to today. However, their undeniable qualities are counterbalanced by the fundamental limitations of handmade representations. In this work we show that we can train a single learnable frontend that outperforms mel-filterbanks on a wide range of audio signals, including speech, music, audio events and animal sounds, providing a general-purpose learned frontend for audio classification. To do so, we introduce a new principled, lightweight, fully learnable architecture that can be used as a drop-in replacement of mel-filterbanks. Our system learns all operations of audio features extraction, from filtering to pooling, compression and normalization, and can be integrated into any neural network at a negligible parameter cost. We perform multi-task training on eight diverse audio classification tasks, and show consistent improvements of our model over mel-filterbanks and previous learnable alternatives. Moreover, our system outperforms the current state-of-the-art learnable frontend on Audioset, with orders of magnitude fewer parameters.

For agents

The same record, over MCP at https://syntology.ai/mcp:

get_harvested_code_for_paper("2101.08596")
get_code_for_paper("2101.08596")
have("2101.08596")

The run record, dated, one paper per request, free:

curl https://syntology.ai/api/ran/2101.08596.json

A badge for a README (the split and the date, never a ratio):

[![Syntology run record](https://syntology.ai/api/ran/2101.08596.svg)](https://syntology.ai/paper/2101.08596)

Connect an agent — have() is free.